High speed 3D photomechanics testing via additional temporal sampling

Thumbnail Image
Suresh, Vignesh
Major Professor
Li, Beiwen
Bentil, Sarah A
Chandra, Abhijit
Oliver, James H
Winer, Eliot
Committee Member
Journal Title
Journal ISSN
Volume Title
Research Projects
Organizational Units
Journal Issue
Is Version Of
Mechanical Engineering
The main objective of measurement is to give the users a better perception of the three-dimensional objects around us. Three-dimensional (3D) shape measurement aids this process as the users are able to get a better sense of the object with the depth information. 3D shape measurement has been widely applied in many fields such as automobile, entertainment, manufacturing, and medicine. The measurement methods can be broadly classified into two categories: contact and non-contact-based techniques. Though the contact methods (such as coordinate measuring machines) are accurate, they are subjected to the risk of damaging the surface of the sample during measurement. The non-contact methods (such as time of flight, stereo vision, and structured light system) were invented to address this limitation by performing a non-destructive evaluation. However, it has always been a challenging task to perform accurate measurements of objects moving at high speeds. Over the years, researchers have made significant progress in this direction. The binary defocusing method is one such advancement, which involves generating quasi-sinusoidal patterns by defocusing 1-bit binary patterns. With the help of this method, it has been possible to perform super fast measurements (e.g., kHz) with high spatial resolutions. Despite the speed breakthrough, there are a number of challenges associated with such technology: (1) motion-induced errors such as artifacts are present in a measurement scene if the object moves at a high- speed; (2) it is a challenging task to perform accurate 3D shape measurement for objects with a highly reflective surface; (3) it is difficult to perform accurate subsequent analysis (e.g., strain, displacement, etc.) of highly reflective objects subjected to loading; (4) it is hard to perform mechanical analysis (e.g., displacement, strain, etc.) for objects that cannot be speckle painted; and (5) it is difficult to perform high-accuracy 3D shape measurements at high-speeds. The first challenge is common in dynamic measurements, especially when the sampling rate is less than the speed at which the object moves. Upgrading the hardware capabilities is always a typical solution, but it is always desired to avoid the increase in hardware cost and have control using the software. We developed a novel software approach for alleviating the errors induced by the object motion in 3D shape measurement. The approach consisted of taking advantage of the additional temporal sampling by capturing two images instead of one in each pattern illumination cycle. Using this additional sampling (capturing two images in one cycle), we developed an iterative approach to reduce the motion-induced errors. Simulation and experiment illustrated the success of our method. In particular, we tested the method by measuring the free- swinging of a spherical ball, for which the motion-induced errors were reduced by 90 % by using our method. The second challenge is typical while measuring objects with high reflectivity. Highly reflective objects tend to have saturation while subjected to capture using a camera. Employing multiple cameras and capturing images of the object from different orientations might be a potential solution. However, the data fusion from multiple optical systems might be difficult. We developed a novel high-dynamic-range method to avoid the saturation problem in imaging objects with high reflectivity. This method also takes advantage of the additional temporal sampling by capturing two images with different brightness levels in one projector cycle. Experiments demonstrate the success of the proposed method by performing an accurate 3D shape measurement of a highly reflective steel plate and a highly reflective metal fan. Addressing the above-mentioned challenge has enabled us to perform high-speed high-dynamic-range 3D shape measurements. Under such a platform, we are attempting to introduce this solution to a different field. Digital image correlation (DIC) is a non- contact method for estimating strain and displacement of objects subjected to deformation. Images are captured before and after deformation, and the displacement data is obtained by correlation analysis. However, the captured camera image tends to have a bright saturated spot when objects with high reflectivity are illuminated with a light source. In such saturated regions, the displacement information cannot be obtained. We extend the developed approach of additional temporal sampling to perform high-dynamic-range digital image correlation (HDR DIC). Two images with different brightness levels are captured in one cycle, and the final unsaturated image is obtaining by combining the two images. The unsaturated image is used for DIC to estimate a full field displacement map. Experiments demonstrate the success of the proposed HDR DIC method by accurately estimating the displacement of a highly reflective steel plate. The above-mentioned challenge was about performing mechanical analysis using a well-established Digital Image Correlation (DIC) method. DIC involves spraying speckle patterns on the objects to estimate the displacement. However, in the case of certain objects, such as electric circuit boards, it might not be possible to spray the speckles as it might affect the functionality of the object. To address this problem, we developed a new method called Digital Height Correlation (DHC) to measure the deformation of the object. The main advantage of this method is that it does not require spraying of speckle patterns for measuring the surface deformation. The DHC method involves measuring the 3D shape of the object and extracting the height data. This height data is used to estimate the surface deformation. The 3D profile of the object is obtained before and after the displacement of the object. The corresponding height maps are extracted and are used to obtain the displacement map. Experiments demonstrate the success of the proposed DHC method in measuring surface deformations. The additional temporal sampling has been effective in solving the problems associated with 3D shape measurement, such as saturated objects, objects in motion, etc. However, it is still a multi-shot method (involves capturing multiple images), and it might not work well while imaging an object moving at extremely high-speeds (in the order of kHz). The Fourier transform method in fringe projection technique can be used in such scenarios because of its single-shot nature. However, the Fourier transform method compromises accuracy; the phase map (estimated using the Fourier method) has several artifacts and noise around the boundary region of the object. Therefore, the 3D surface reconstructed from this phase map has artifacts. We developed a deep learning based method called Phase Map Enhancement Net (PMENet) for improving the accuracy of the Fourier transform method. The PMENet is trained to predict a high-quality phase map from the Fourier transform phase map. During the training stage, the performance of the PMENet was evaluated using the phase maps obtained from an N-step phase-shifting method. The training dataset was generated using a virtual 3D scanning system. Experimental results demonstrated that the PMENet could successfully enhance the quality of the 3D geometries reconstructed by the Fourier transform method as the mean and root-mean-square-error reduced by 66 % and 43 %, respectively. In summary, this dissertation significantly advances the research of high-speed and high-accuracy 3D shape measurement. Using additional temporal sampling, we can solve the challenges (such as motion-induced errors and high-dynamic-range imaging) associated with 3D shape measurement, and can also achieve accurate mechanical analysis (HDR DIC). The DHC method addresses the problem of performing accurate mechanical analysis for objects that cannot be spray painted. The PMENet enhances the accuracy of the Fourier transform phase map, thereby enabling simultaneous high-speed and high accuracy 3D shape measurement. Contributions to this research can potentially benefit various fields in both industry and academia, where both speed and accuracy are the major concerns.
Subject Categories